Wide band video signal denoiser and method for denoising

Abstract

A method and apparatus for high speed denoising of an input signal received by a radar warning receiver (RWR) system. The novel system digitizes the input signal into sample windows each having a number of sample data points. The novel system includes a signal transformation processor system which convolves Haar wavelet basis functions across individual time segments of the input signal to produce sets of coefficients. Signal noise is poorly represented by the selected Haar wavelet basis functions while expected pulse signal characteristics are well represented by the basis functions. Within the wavelet transformation process, a plurality of frequency resolution levels are simultaneously determined, each frequency resolution level having a fixed number of correlation coefficients equal to the number of sample points in the sample window. Each coefficient also corresponds to a different basis function, or a same basis function applied to a different time segment. Each level contains a different number of represented frequency bands and a different number of time segments per band. For each level, the level's coefficients are sorted into coefficient maps and the largest coefficients are selected from each map. The coefficient map best packaging the input signal energy is then selected. The coefficients of the best coefficient map can be used to (1) reconstruct a denoised digital or analog signal using a re-transformation for application to a video pulse processor for threat characterization or (2) to produce a digital report indicating vital pulse information including leading edge pulse position, pulse amplitude, pulse duration, etc.

What is claimed is: 
     
       1. In a radar warning receiver, a method for denoising an input wide band video signal window, said method comprising the steps of: a) selecting a plurality of wavelet basis functions that represent expected signal waveforms of said input wide band video signal window, each of said plurality of wavelet basis functions composed of values of either "1" or "-1";   b) performing high speed real-time convolution of said input wide band video signal window against said plurality of wavelet basis functions to generate a plurality of correlation coefficients wherein groups of correlation coefficients share a common frequency resolution level, said step b) of performing high speed convolution being performed without multiplication functions to determine said plurality of correlation coefficients;   c) thresholding said plurality of correlation coefficient to select those correlation coefficients that most efficiently package the energy of the input wide band video signal window;   d) discarding correlation coefficients not thresholded by step c); and   e) generating a denoised version of said input wide band video signal window by performing a high speed real-time wavelet reconstruction based on correlation coefficients thresholded by step c).   
     
     
       2. A method as described in claim 1 wherein said plurality of wavelet basis functions are Haar wavelet basis functions and wherein said high speed real-time convolution step b) operates based on an input wide band video signal window signal rate of 64 MHz. 
     
     
       3. A method for denoising an input digital signal, said method comprising the computer implemented steps of: a) sampling said input digital signal in sample windows, each sample window comprising x discrete sample points;   b) convolving said sample window with selected wavelet basis functions across n frequency resolution levels to generate n sets of x correlation coefficients, each frequency resolution level generating x correlation coefficients wherein each frequency resolution level, r, comprises 2 r  separate frequency bands, each frequency band associated with a wavelet basis function, and wherein said step b) further comprises the step of performing n separate frequency resolution level convolution processes each convolution process of a level, r, convolving 2 r  wavelet basis functions with discrete sample points of said sample window to generate a set of x correlation coefficients for said level, r:   c) generating a set of n coefficient maps, one coefficient map per frequency resolution level of said n frequency resolution levels;   d) determining a best coefficient map of said n coefficient maps that most efficiently represents said x discrete sample points of said sample window; and   e) recording selected correlation coefficients of said best coefficient map and discarding all remaining correlation coefficients.   
     
     
       4. A method as described in claim 3 further comprising the step of performing a re-transformation of said selected correlation coefficients of said best coefficient map to reconstruct an output denoised digital signal. 
     
     
       5. A method as described in claim 4 further comprising the steps of: normalizing said output denoised digital signal to produce a normalized output denoised digital signal; and   converting said normalized output denoised digital signal into an analog denoised signal.   
     
     
       6. A method as described in claim 3 further comprising the step of generating a digital report based on said selected correlation coefficients, said digital report including a leading edge pulse time and a pulse peak amplitude of an existing pulse within said sample window. 
     
     
       7. A method as described in claim 3 wherein each frequency resolution level convolution process of a level, r, convolves said 2 r  wavelet basis functions across x/2 r  time segments of said sample window to generate said set of x correlation coefficients for said level, r. 
     
     
       8. A method as described in claim 3 wherein said step c) comprises the steps of: ranking said set of x correlation coefficients for each frequency resolution level in order of magnitude to produce said set of n coefficient maps; and   for each coefficient map of said n coefficient maps, determining correlation coefficients that fall within a determined threshold, wherein said selected coefficients are selected by said above step of determining correlation coefficients that fall within said determined threshold.   
     
     
       9. A method as described in claim 3 wherein said step d) comprises the steps of: determining a drop-off value of coefficient magnitude for each of said n coefficient maps; and   selecting said best coefficient map as that coefficient map of said n coefficient maps having a sharpest drop-off value.   
     
     
       10. A method as described in claim 3 wherein x is equal to or less than 32 and n is equal to or less than 5. 
     
     
       11. In a radar warning receiver (RWR), a method for denoising an input analog signal, said method comprising the steps of: a) converting said input analog signal into an input digital signal, said input digital signal having sample windows, each sample window comprising x discrete sample points;   b) convolving said sample window with selected wavelet basis functions across n frequency resolution levels to generate n sets of x correlation coefficients, each frequency resolution level generating x correlation coefficients wherein each frequency resolution level, r, comprises 2 r  separate frequency bands, each frequency band associated with a wavelet basis function and wherein said step b) further comprises the step of performing n separate frequency resolution level convolution processes, each convolution process of a level, r, convolving 2 r  wavelet basis functions with discrete sample points of said sample window to generate a set of x correlation coefficients for said level, r;   c) generating a set of n coefficient maps, one coefficient map for each frequency resolution level of said n frequency resolution levels;   d) determining a best coefficient map of said n coefficient maps that most efficiently represents said x discrete sample points of said sample window and determining selected correlation coefficients of said best coefficient map;   e) performing a re-transformation of solely said selected correlation coefficients of said best coefficient map to reconstruct an output denoised digital signal; and   f) converting said output denoised digital signal into an output denoised analog signal.   
     
     
       12. A method as described in claim 11 further comprising the step of applying said output denoised analog signal to an RWR video pulse processor for threat characterization. 
     
     
       13. A method as described in claim 11 where said step f) comprises the steps of: normalizing said output denoised digital signal to produce a normalized output denoised digital signal; and   using a digital to analog converter circuit to convert said normalized output denoised digital signal into said output denoised analog signal.   
     
     
       14. A method as described in claim 11 wherein each frequency resolution level convolution process of a level, r, convolves said 2 r  wavelet basis functions across x/2 r  time segments of said sample window to generate said set of x correlation coefficients for said level, r. 
     
     
       15. A method as described in claim 11 wherein said step c) comprises the steps of: ranking said set of x correlation coefficients for each frequency resolution level in order of magnitude to produce said set of n coefficient maps; and   for each coefficient map of said n coefficient maps, determining correlation coefficients that fall within a determined threshold, wherein said selected coefficients are selected by said above step of determining correlation coefficients that fall within said determined threshold.   
     
     
       16. A method as described in claim 11 wherein said step d) comprises the steps of: determining a drop-off value of coefficient magnitude for each of said n coefficient maps; and   selecting said best coefficient map as that coefficient map of said n coefficient maps having a sharpest drop-off value.   
     
     
       17. A radar warning receiver (RWR) system comprising: an wide band receiver for receiving an analog signal;   an analog to digital converter circuit for converting said analog signal into an input digital signal;   a processor system coupled to receive said input digital signal in sample windows, each sample window comprising x discrete sample points, said processor system for performing the steps of: a) convolving said sample window with selected wavelet basis functions across n frequency resolution levels to generate n sets of x correlation coefficients, each frequency resolution level generating x correlation coefficients wherein each frequency resolution level, r, comprises 2 r  separate frequency bands, each frequency band associated with a wavelet basis function and wherein said step a) further comprises the step of performing n separate frequency resolution level convolution processes, each convolution process of a level, r, convolving 2 r  wavelet basis functions with discrete sample points of said sample window to generate a set of x correlation coefficients of said level, r;   b) generating a set of n coefficient maps, one coefficient map for each frequency resolution level of said n frequency resolution levels;   c) determining a best coefficient map of said n coefficient maps that most efficiently represents said x discrete sample points of said sample window and determining selected correlation coefficients of said best coefficient map; and   d) performing a re-transformation of solely said selected correlation coefficients of said best coefficient map to reconstruct an output denoised digital signal; and     a digital to analog converter coupled to said processor system for converting said output denoised digital signal into an output denoised analog signal.   
     
     
       18. An RWR system as described in claim 17 further comprising an RWR pulse processor coupled to receive said output denoised analog signal, said RWR pulse processor for performing threat characterization. 
     
     
       19. An RWR system as described in claim 17 where said step d) comprises the step of normalizing said output denoised digital signal. 
     
     
       20. An RWR system as described in claim 17 wherein each frequency resolution level convolution process of a level, r, convolves said 2 r  wavelet basis functions across x/2 r  time segments of said sample window to generate said set of x correlation coefficients of said level, r. 
     
     
       21. An RWR system-as described in claim 17 wherein said step b) comprises the steps of: ranking said set of x correlation coefficients for each frequency resolution level in order of magnitude to produce said set of n coefficient maps; and   for each coefficient map of said n coefficient maps, determining correlation coefficients that fall within a determined threshold, wherein said selected coefficients are selected by said above step of determining correlation coefficients that fall within said determined threshold.   
     
     
       22. An RWR system as described in claim 17 wherein said step c) comprises the steps of: determining a drop-off value of coefficient magnitude for each of said n coefficient maps; and   selecting said best coefficient map as that coefficient map of said n coefficient maps having a sharpest drop-off value.   
     
     
       23. A computer system for denoising an input digital signal, said computer system comprising a processor coupled to a bus and a computer readable memory unit coupled to said bus, said computer readable memory unit containing a program for causing said processor to perform the steps of: sampling said input digital signal into sample windows, each sample window comprising x discrete sample points;   b) convolving said sample window with selected wavelet basis functions across n frequency resolution levels to generate n sets of x correlation coefficients, each frequency resolution level generating x correlation coefficients where each frequency resolution level, r, comprises 2 r  separate frequency bands, each frequency band associated with a respective wavelet basis function and wherein said step b) further comprises the step of performing n separate frequency resolution level convolution processes, each convolution process of a level, r, convolving 2 r  wavelet basis functions with discrete sample points of said sample window to generate a set of x correlation coefficients of said level, r;   c) generating a set of n coefficient maps, one coefficient map for each frequency resolution level of said n frequency resolution levels;   d) determining a best coefficient map of said n coefficient maps that most efficiently represents said x discrete sample points of said sample window; and   e) recording selected correlation coefficients of said best coefficient map.   
     
     
       24. A system as described in claim 23 wherein said processor further performs the step of performing a re-transformation of said selected correlation coefficients of said best coefficient map to reconstruct an output denoised digital signal. 
     
     
       25. A system as described in claim 24 wherein said processor further performs the step of normalizing said output denoised digital signal to produce a normalized output denoised digital signal; and further comprising a digital to analog converter circuit for converting said normalized output denoised digital signal into an output denoised analog signal.   
     
     
       26. A system as described in claim 23 wherein said processor further performs the step of generating a digital report based on said selected correlation coefficients, said digital report including a leading edge pulse time and a pulse peak amplitude. 
     
     
       27. A system as described in claim 23 wherein said each frequency resolution level convolution process of a level, r, convolves said 2 r  wavelet basis functions across x/2 r  time segments of said sample window to generate said set of x correlation coefficients of said level, r. 
     
     
       28. A system as described in claim 23 wherein said step c) comprises the steps of: ranking said set of x correlation coefficients for each frequency resolution level in order of magnitude to produce said set of n coefficient maps; and   for each coefficient map of said n coefficient maps, determining correlation coefficients that fall within a determined threshold, wherein said selected coefficients are selected by said above step of determining correlation coefficients that fall within said determined threshold.   
     
     
       29. A system as described in claim 23 wherein said step d) comprises the steps of: determining a drop-off value of coefficient magnitude for each of said n coefficient maps; and   selecting said best coefficient map as that coefficient map of said n coefficient maps having a sharpest drop-off value.